Catapult devices are GPS trackers worn by athletes. Wearable-based tracking technologies are used throughout sport to support performance monitoring. In addition to GPS capability, these devices contain inertial sensors comprising of an accelerometer (to measure acceleration forces), a gyroscope (to measure rotation), and a magnetometer (to measure body orientation). Inertial sensors collect data in three axes, or directions, allowing sensitive ‘maps’ of athlete movements and actions to be created. For Catapult’s website, they claim: “The combination of the wearable tracking device and the inertial sensors creates a powerful athlete monitoring tool that ensures that key performance decisions are always supported with objective data.” The sports performance department at LA Galaxy uses Catapult data to make decisions about performance readiness, rehabilitation, and training prescription.
This data is Catapult data collected over the course of the U17 2022-2023 season. I will specifically be look at U17 game data for that season. As a student in the USC Sports Science program and an intern at the LA Galaxy Sports Performance Department, I’ve had the opportunity to assist with collecting this data since January 2023. This data is typically visualized using either Catapult’s Cloud where they offer many widgets to visualize data, or an internal athlete management system. LA Galaxy has been developing an athlete management system using Microsoft Azure. They export the data from Catapult and import it to Azure and have customized many different dashboards to visualize data. For this project, I decided to export CSVs directly from Catapult and try to wrangle the data myself.
Catapult data is collected at every training session and game. The players wear the devices on vests produced by Catapults and the GPS units are stored in a secure pouch on the back of the vest. During training or games, a member of the sport performance department will have an iPad which has the Vector app created by Catapult. The Vector app allows the user to input information about the training session or game, and it produces a live view of the Catapult data per player. The user can start and stop “Periods” based on training drills and which players are participating. After training, all the units are collected, put into a dock, and uploaded to a computer. This data is then available in the Cloud and can be exported to the athlete management system for further visualization.
When thinking about this data, my research question became: does fatigue affect player’s physical performance in soccer matches? More specifically, are players less physically productive when they are tired? To answer this question, I looked at the data at a few levels. To start off, I look at the difference in player’s maximum velocities in the first half of games vs the second half. Then, I look at a string of five games in seven games that the team played in difficult conditions in the MLS Next Tournament, which was played in June of this year.
Methods
Loading the necessary packages
library(dplyr)
Attaching package: 'dplyr'
The following objects are masked from 'package:stats':
filter, lag
The following objects are masked from 'package:base':
intersect, setdiff, setequal, union
The following object is masked from 'package:dplyr':
group_rows
# This is where I stored the downloaded CSVs from Catapultsetwd("/Users/sylwialipior/Downloads/pm566-01-lab/U17 2022-2023 Data")# Get a list of all CSV files containing "_GD_"filenames <-list.files(pattern ="_GD_.*\\.csv$")
Preparing the data frame
When you export bulk CSVs from Catapult, you get observations for every player involved in the session for 1699 variables. A lot of that data is a little redundant, but I wrote a function to subset the data with only around 34 variables of interest to make it more manageable. The CSVs don’t have the activity name easily accessible, so I wrote a function to extract the names from the names of the CSV files. I then wrote a for loop to read in all the data (~57 CSVs which corresponds to data from 57 games), making a new variable for the date of the session, and a new variable for the activity name. Finally, I de-identified the data since the data contains player names.
setwd("/Users/sylwialipior/Downloads/pm566-01-lab/U17 2022-2023 Data")# List of desired variablesdesired_variables <-c("Player.Name", "Period.Name", "Period.Number", "Position.Name", "Total.Duration", "Total.Distance", "Total.Player.Load", "Player.Load.Per.Minute", "Player.Load.Per.Metre", "Meterage.Per.Minute", "Maximum.Velocity", "High.Speed.Distance.12mph.14mph", "Very.High.Speed.Distance.14mph.17mph", "Sprinting.Distance.17.19mph", "Supra.Max.Velocity..19mph", "Number.of.Sprints", "Velocity.Band.7.Average.Effort.Count", "Velocity.Band.8.Average.Effort.Count", "Max.Vel....Max.", "Profile.Max.Velocity", "Explosive.Efforts", "HSD.min", "Total.High.Intensity.Bouts..THIB.", "Maximal.High.Intensity.Bouts..MHIB.", "Accels..2.5...3.m.s.s.", "Accels..3...3.5.m.s.s.", "Accels....3.5.m.s.s.", "Decels...2.5....3.m.s.s", "Decels...3....3.5.m.s.s.", "Decels.....3.5.m.s.s.", "Acceleration.Density", "Acceleration.Density.Index")#Function to read only columns of interestread_selected_columns <-function(filename) {# Read the entire CSV data <-read.csv(filename, skip =9, header =TRUE, sep =",")# Subset the data to keep only the desired columns data <- data[, desired_variables, drop =FALSE]return(data)}# Function to extract and format the activity name from filenameextract_activity_name <-function(filename) {# Extract the part of the filename after U17 and before the file extension name_part <-sub(".*U17_([^\\.]+)\\.csv$", "\\1", filename)# Replace underscores with spaces activity_name <-gsub("_", " ", name_part)# Prepend "U17 " to the modified namepaste("U17", activity_name)}# Initialize an empty list to hold individual data framesdata_frames <-list()# Loop through each file, read it, and add to the listfor (filename in filenames) { df <-read_selected_columns(filename)#Extract the activity name from the filename activity_name <-extract_activity_name(filename)# Extract date information from the filename date_string <-substr(filename, 1, 10) # Assuming the date is always the first 10 characters date_obj <-as.Date(date_string, format ="%Y_%m_%d")# Add new columns for date and activity name df$Date <- date_obj df$Activity_Name <- activity_name data_frames[[filename]] <- df # using filename as list name just for clarity, can use any naming convention}# Combine all data frames into one master data framemaster_df <-bind_rows(data_frames)## De-identifying the data# Generate a unique identifier for each player nameunique_players <-unique(master_df$Player.Name)name_mapping <-data.frame(Original_Name = unique_players,Identifier =paste0("Player_", seq_along(unique_players)))# Replace the actual player names with the generated identifiersmaster_df$Player.Name <- name_mapping$Identifier[match(master_df$Player.Name, name_mapping$Original_Name)]
Initial visualization of the data
Next, I wanted to make sure the data looks how I would expect it to. Since the session names are inputted by staff, there is some room for error, and I wanted to make sure I have only game data here. To accomplish this, I decided to plot maximum velocity for each activity by month.
# Extract month and year from the Date column to create a new 'Month' columnmaster_df$Month <-format(master_df$Date, "%m")# List of unique monthsunique_months <-unique(master_df$Month)# # Loop through each month, create a plot, and then add a page break# for (month in unique_months) {# sub_df <- master_df[master_df$Month == month, ]# # print(# ggplot(sub_df, aes(x = Activity_Name, y = Maximum.Velocity, fill = Activity_Name)) +# geom_boxplot() +# theme(axis.text.x = element_text(angle = 45, hjust = 1)) +# labs(title = paste("Max Speeds of Players for Month", month),# x = "Activity",# y = "Max Speed (mph)") + # theme(legend.position = "none")# )# # # Add a page break after each plot# cat("\\newpage")# }# Filter out the unwanted activitiesmaster_df <- master_df %>%filter(!(Activity_Name %in%c("U17 GD vs RSL", "U17 GD Pre Season Day 2")))
From looking at the initial box plots, I figured out that there is some data in the set that doesn’t belong. Specifically, “U17 GD vs RSL” seems to have an issue with the GPS data since the maximum velocities are so low, so I decided to remove that data. In addition, the maximum velocities for “U17 GD Pre Season Day 2” are much lower than expected. Once I looked at the period names, I realized that this is data from a training session that was mislabeled as a game, so I removed it from the data set. I included the code to produce these box plots, but decided not to render here, because I did not use them for further analysis.
Player Load Scatter Plots
For some more initial visualization, I decided to look at the relationship between “Player Load”, which is defined by Catapult as “the sum of the accelerations across all axes of the internal tri-axial accelerometer during movement”, and a few other physical metrics. Specifically, I looked at scatterplot of Player Load vs Total Distance Covered, Total Number of Sprints, Explosive Efforts, and Total High Intensity Bouts.
library(ggplot2)library(dplyr)library(knitr)# Filter data for when Period.Name is "Session" and for games in Septemberfiltered_df <- master_df %>%filter(Period.Name =="Session")ggplot(filtered_df, aes(x = Total.Player.Load, y = Total.Distance, color = Player.Name)) +geom_point() +labs(title ="Scatterplot of Player Load vs. Distance Covered", x ="Total Player Load",y ="Total Distance") +theme(legend.position ="none")
ggplot(filtered_df, aes(x = Total.Player.Load, y = Number.of.Sprints, color = Player.Name)) +geom_point() +labs(title ="Scatterplot of Player Load vs. Total Number of Sprints", x ="Total Player Load",y ="Total Number of Sprints") +theme(legend.position ="none")
ggplot(filtered_df, aes(x = Total.Player.Load, y = Explosive.Efforts, color = Player.Name)) +geom_point() +labs(title ="Scatterplot of Player Load vs. Explosive Efforts", x ="Total Player Load",y ="Explosive Efforts") +theme(legend.position ="none")
ggplot(filtered_df, aes(x = Total.Player.Load, y = Total.High.Intensity.Bouts..THIB., color = Player.Name)) +geom_point() +labs(title ="Scatterplot of Player Load vs. Total High Intensity Bouts", x ="Total Player Load",y ="Total High Intensity Bouts") +theme(legend.position ="none")
After looking at the plots, as expected there is a strong positive correlation between Player Load and all the other physical metrics. In other words, as a player covers more distance, or performs more sprints, explosive efforts, or high intensity bouts, their player load is expected to be higher.
Maximum Velocity Analysis
I thought it would be interesting to analyze maximum velocity from the dataset. Unfortunately, the data is a little difficult, because there are various Period Names that could signify either First Half or Second Half game data. Therefore, I used mutate to add a variable called “Period.Name.Halves” to denote which observations are from the first half vs the second half. I also removed Goal Keeper data, since their data looks very different from field players due to the nature of their position. I added a threshold of 10mph for speed, to make sure that I am actually using game data, and not some other potentially mislabeled data. I then visualized the maximum velocity data in a few different ways.
library(dplyr)library(tidyr)library(ggplot2)# Mutate new variable based on criteriamaster_df <- master_df %>%mutate(Period.Name.Halves =case_when( Period.Name %in%c("1st Half", "0- 10 min", "10-45min", "0-15mins", "15-30mins", "30-45mins") ~"First Half", Period.Name %in%c("2nd Half", "45-60mins", "60-70mins", "70-90mins", "45-70mins", "70-75mins", "75-90mins", "66-80min", "80-85min", "85-90min", "45-50mins", "50-65mins", "60-75mins","45-75mins", "75-83min", "83-90min", "45-55mins", "55-60mins", "60-72mins", "72-90mins","56-60mins", "55-70mins", "70-77mins", "77-90mins", "70-74mins", "74-83mins", "83-90mins","60-69mins", "69-77mins", "77-83mins", "75-78mins", "78-85mins", "85-90mins", "75-80mins","80-90mins", "45-58mins", "58-75mins", "70-85mins", "60-65mins", "60-68mins", "68-80mins") ~"Second Half",TRUE~ Period.Name # keeps original period names for the rest ))# Filter the dataset to retain only the maximum velocity observation per player, per activity, and per halffiltered_max_speed_df <- master_df %>%# Remove Goal Keeper data since they would skew the datafilter(Position.Name !="Goal Keeper") %>%# Retain only observations where Maximum.Velocity is at least 10filter(Maximum.Velocity >=10) %>%group_by(Player.Name, Activity_Name, Period.Name.Halves) %>%filter(Maximum.Velocity ==max(Maximum.Velocity, na.rm =TRUE)) %>%ungroup()# Filter data to include only rows where Period.Name.Halves is "First Half" or "Second Half"filtered_df_halves <- filtered_max_speed_df %>%filter(Period.Name.Halves %in%c("First Half", "Second Half"))# Boxplot comparing max speed between the two halvesggplot(filtered_df_halves, aes(x = Period.Name.Halves, y = Maximum.Velocity, fill = Period.Name.Halves)) +geom_boxplot() +labs(title ="Comparison of Max Speed in First Half vs Second Half",x ="Half",y ="Max Speed (mph)") +theme(legend.position ="none")
# Group by Activity and Period.Name.Halves, then find the top 3 players by maximum velocitytop_players_by_activity <- filtered_df_halves %>%group_by(Activity_Name, Period.Name.Halves) %>%top_n(3, Maximum.Velocity) %>%ungroup()# Count the number of times each player is in the top 3 for the first halfplayer_counts_first_half <- top_players_by_activity %>%filter(Period.Name.Halves =="First Half") %>%# filter for "First Half" onlygroup_by(Player.Name, Period.Name.Halves) %>%summarise(Times_in_Top_3 =n()) %>%arrange(desc(Times_in_Top_3), Player.Name)
`summarise()` has grouped output by 'Player.Name'. You can override using the
`.groups` argument.
# Count the number of times each player is in the top 3 for the first halfplayer_counts_second_half <- top_players_by_activity %>%filter(Period.Name.Halves =="Second Half") %>%group_by(Player.Name, Period.Name.Halves) %>%summarise(Times_in_Top_3 =n()) %>%arrange(desc(Times_in_Top_3), Player.Name)
`summarise()` has grouped output by 'Player.Name'. You can override using the
`.groups` argument.
print(player_counts_first_half)
# A tibble: 24 × 3
# Groups: Player.Name [24]
Player.Name Period.Name.Halves Times_in_Top_3
<chr> <chr> <int>
1 Player_14 First Half 22
2 Player_17 First Half 17
3 Player_3 First Half 14
4 Player_4 First Half 11
5 Player_7 First Half 11
6 Player_19 First Half 9
7 Player_12 First Half 8
8 Player_15 First Half 7
9 Player_10 First Half 6
10 Player_20 First Half 6
# ℹ 14 more rows
print(player_counts_second_half)
# A tibble: 24 × 3
# Groups: Player.Name [24]
Player.Name Period.Name.Halves Times_in_Top_3
<chr> <chr> <int>
1 Player_14 Second Half 19
2 Player_4 Second Half 14
3 Player_12 Second Half 12
4 Player_17 Second Half 12
5 Player_7 Second Half 12
6 Player_19 Second Half 11
7 Player_21 Second Half 8
8 Player_26 Second Half 7
9 Player_3 Second Half 7
10 Player_11 Second Half 6
# ℹ 14 more rows
# Extract top 5 player names for each halftop_5_names_first <- player_counts_first_half$Player.Name[1:5]top_5_names_second <- player_counts_second_half$Player.Name[1:5]# Combine and get unique namestop_5_names_combined <-unique(c(top_5_names_first, top_5_names_second))# Filter original datafiltered_first_half <- player_counts_first_half %>%filter(Player.Name %in% top_5_names_first)filtered_second_half <- player_counts_second_half %>%filter(Player.Name %in% top_5_names_second)# Combine the datacombined_filtered_data <-rbind(filtered_first_half, filtered_second_half)# Create the bar chartggplot(combined_filtered_data, aes(x = Period.Name.Halves, y = Times_in_Top_3, fill = Player.Name)) +geom_bar(stat ="identity", position ="stack") +labs(title ="Top 5 Players Who Appeared in Top 3 Max Speeds by Half",x ="Half",y ="Times in Top 3") +theme_minimal()
# Extract top speeds for each player and each halftop_speeds_by_half <- filtered_df_halves %>%group_by(Player.Name) %>%summarise(`First Half Speed`=max(Maximum.Velocity[Period.Name.Halves =="First Half"], na.rm =TRUE),`Second Half Speed`=max(Maximum.Velocity[Period.Name.Halves =="Second Half"], na.rm =TRUE) )melted_data <- top_speeds_by_half %>%gather(key ="Half", value ="Speed", `First Half Speed`, `Second Half Speed`)# Spaghetti plot of best first half speed and best second half speed by playerggplot(melted_data, aes(x = Half, y = Speed, group = Player.Name, color = Player.Name)) +geom_line(size =0.5) +geom_point(size =3) +labs(title ="Top Observed Velocity: First Half vs Second Half",x ="Game Half",y ="Top Velocity") +theme(legend.position ="none")
Warning: Using `size` aesthetic for lines was deprecated in ggplot2 3.4.0.
ℹ Please use `linewidth` instead.
`summarise()` has grouped output by 'Player.Name'. You can override using the
`.groups` argument.
melted_avg_data <- avg_speed_by_half %>%pivot_longer(cols = Average.Velocity, names_to ="Attribute", values_to ="Speed")# Create the spaghetti plotggplot(melted_avg_data, aes(x = Period.Name.Halves, y = Speed, group = Player.Name, color = Player.Name)) +geom_line(size =0.5) +geom_point(size =3) +labs(title ="Average Observed Velocity: First Half vs Second Half",x ="Game Half",y ="Average Velocity") +theme(legend.position ="none")
The “Comparison of Max Speed in First Half vs Second Half” box plot showed that there doesn’t seem to be a big difference in the average speed of players in the first half and second half. Something I could consider is removing players that did not play a full game from the data, since players are often subbed on in the second half. These substitutes could be running faster since they aren’t fatigued yet, so they could be increasing the average. There seems to be a larger range of speed in the second half, which makes sense since players are getting fatigued.
Next, I was interested in looking at the individual player level. I took the frequency of players that appeared in the top 3 max speed values per activity. “Player_14” is clearly the fastest player, since he appears in the top 3 the most times out of anyone. I made a stacked box plot looking at the five players that appeared in the top 3 for maximum velocity the most number of times. Four of the players appear in the top 3 the most times for both the first half and second half, but player 3 appeared in the top 3 the third most times for first half and player 12 appeared in the top 3 the third most times for the second half. That tells me that player 12 might be a second half substitute often while player 3 gets subbed off in the second half.
I then made a spaghetti plot looking at the top recorded velocity for the first half vs second half for each player to see if there is a difference in performance between halves. Interestingly, the trend seems to be that players achieve higher maximum velocities in the second half of games. That makes sense, because player’s might be experience fatigue and might make errors where they have to achieve very high speeds to deal with counter attacks. I made a similar spaghetti plot, except with average velocity across all activities. Now, the trend seems to be the opposite.
Extract Games from MLS Next Tournament (June 2023)
In June of 2023, the LA Galaxy U17 team won the MLS Next Tournament. In order to hoist the trophy, they played 5 matches in 7 days in very difficult and humid conditions in Dallas, Texas. To perform an analysis on this data, I had to subset the master data frame to extract the data for these matches. Then, I calculated the match totals for the following physical metrics: total distance, total high intensity bouts, total player load, total explosive efforts, total sprints, total high speed distance (12-14 mph), total very high speed distance (14-17 mph), and total sprinting distance (17-19 mph). I created a table summarizing this data, and made multiple bar charts.
library(dplyr)library(stringr)library(ggplot2)filtered_df_2023_06 <- master_df %>%# Remove Goal Keeper data since they would skew the datafilter(Position.Name !="Goal Keeper") %>%filter(str_detect(Date, "^2023-06"))# Filtering and aggregating datateam_metrics_by_game <- filtered_df_2023_06 %>%filter(Period.Name.Halves %in%c("First Half", "Second Half")) %>%group_by(Activity_Name) %>%summarise(Game_Date =min(Date), # Assuming 'Activity_Date' contains the date of the gameTotal_Distance =sum(Total.Distance, na.rm =TRUE),Total_High_Intensity_Bouts =sum(Total.High.Intensity.Bouts..THIB., na.rm =TRUE),Total_Player_Load =sum(Total.Player.Load, na.rm =TRUE),Total_Explosive_Efforts =sum(Explosive.Efforts, na.rm =TRUE),Total_Sprints =sum(Number.of.Sprints, na.rm =TRUE),Total_High_Speed_Distance =sum(High.Speed.Distance.12mph.14mph, na.rm =TRUE),Total_Very_High_Speed_Distance =sum(Very.High.Speed.Distance.14mph.17mph, na.rm =TRUE),Total_Sprinting_Distance =sum(Sprinting.Distance.17.19mph, na.rm =TRUE) ) %>%ungroup() %>%arrange(Game_Date) # Arrange data by the game date in chronological orderteam_metrics_by_game$Activity_Name <-factor(team_metrics_by_game$Activity_Name, levels = team_metrics_by_game$Activity_Name[order(team_metrics_by_game$Game_Date)])# # Create a table with all our sums# # Custom column names# custom_colnames <- gsub("_", " ", names(team_metrics_by_game))# # summary_table <- kable(team_metrics_by_game, col.names = custom_colnames) %>%# kable_styling(full_width = F, position = "center",# latex_options = c("striped", "scale_down"))# print(summary_table)# Plotting Total Distanceggplot(team_metrics_by_game, aes(x = Activity_Name, y = Total_Distance, fill = Activity_Name)) +geom_bar(stat ="identity") +scale_fill_brewer(palette ="Set3") +labs(title ="Total Distance Covered for Each Game",x ="Game",y ="Total Distance") +theme_minimal() +theme(legend.position ="none", axis.text.x =element_text(angle =45, hjust =1))
# Plotting Total High Intensity Boutsggplot(team_metrics_by_game, aes(x = Activity_Name, y = Total_High_Intensity_Bouts, fill = Activity_Name)) +geom_bar(stat ="identity") +scale_fill_brewer(palette ="Set3") +labs(title ="Total High Intensity Bouts for Each Game",x ="Game",y ="Total High Intensity Bouts") +theme_minimal() +theme(legend.position ="none", axis.text.x =element_text(angle =45, hjust =1))
# Plotting Total Sprintsggplot(team_metrics_by_game, aes(x = Activity_Name, y = Total_Sprints, fill = Activity_Name)) +geom_bar(stat ="identity") +scale_fill_brewer(palette ="Set3") +labs(title ="Total Sprints for Each Game",x ="Game",y ="Total Sprints") +theme_minimal() +theme(legend.position ="none", axis.text.x =element_text(angle =45, hjust =1))
# Plotting Total Explosive Effortsggplot(team_metrics_by_game, aes(x = Activity_Name, y = Total_Explosive_Efforts, fill = Activity_Name)) +geom_bar(stat ="identity") +scale_fill_brewer(palette ="Set3") +labs(title ="Total Explosive Efforts for Each Game",x ="Game",y ="Total Explosive Efforts") +theme_minimal() +theme(legend.position ="none", axis.text.x =element_text(angle =45, hjust =1))
# Plotting Total Player Load with a different color paletteggplot(team_metrics_by_game, aes(x = Activity_Name, y = Total_Player_Load, fill = Activity_Name)) +geom_bar(stat ="identity") +scale_fill_brewer(palette ="Set3") +scale_fill_brewer(palette ="Set3") +# Change this to apply different paletteslabs(title ="Total Player Load for Each Game",x ="Game",y ="Total Player Load") +theme_minimal() +theme(legend.position ="none", axis.text.x =element_text(angle =45, hjust =1))
Scale for fill is already present.
Adding another scale for fill, which will replace the existing scale.
# Plotting Stacked Bar Chartggplot(team_metrics_by_game, aes(x = Activity_Name)) +geom_bar(aes(y = Total_High_Speed_Distance, fill ="High Speed Distance (12-14mph)"), stat ="identity") +geom_bar(aes(y = Total_Very_High_Speed_Distance, fill ="Very High Speed Distance (14-17mph)"), stat ="identity", position ="stack") +geom_bar(aes(y = Total_Sprinting_Distance, fill ="Sprinting Distance (17-19mph)"), stat ="identity", position ="stack") +labs(title ="Distribution of Speed Ranges for Each Game",x ="Game",y ="Total Distance",fill ="Sprinting Speed") +# Added legend title heretheme_minimal() +theme(axis.text.x =element_text(angle =45, hjust =1)) +scale_fill_manual(values =c("High Speed Distance (12-14mph)"="pink","Very High Speed Distance (14-17mph)"="orange","Sprinting Distance (17-19mph)"="red"))
I included a screenshot of the table I made using Kable because I could not get it to render correctly in the HTML no matter what I did, including changing to R markdown. I kept the code, but commented it out.
When looking at total distance covered, there doesn’t seem to be a decrease as the matches progress. Interestingly, the first match had the most sprints, and almost the highest high speed running distance. This is expected since this was the first match of the tournament. There was a noticeable dip in the second match, but this was played the day after the first match. There seemed to be some affects of fatigue in this match. The high speed running distance was the lowest of the tournament. The players also completed the fewest number of sprints in this match. There is also a dip in the number of sprints in the last match, which seems to be an affect of accumulated fatigue. Overall, all of the physical metrics looked at here are have comparable values. When looking at the distribution of speed ranges for each match, there was a noticeable decrease in high speed running in the second match (the one with the least rest) and the last match, again probably due to accumulated fatigue. The distribution of sprinting speeds seems to be fairly similar between matches. As expected, players cover the most distance at high speed, and a small amount of distance at “very” high speeds and sprinting speed.
Conclusion
From my analysis, I was able to get a better understanding of how fatigue affects physical metrics measured by Catapult devices. In my analysis of maximum velocity on the player level, I found that fatigue within a single match does not seem to affect whether players will hit a high maximum velocity. The average maximum velocity for the team is slightly higher in the first half and in the second half, which is expected. Intra-player differences in maximum velocity are very low between the first half and the second half. Furthermore, in my analysis of the MLS Next Tournament, I found that there was not a large impact of accumulated fatigue. The distribution of different sprinting speed distances was similar between the games (i.e. player’s did not seem to be sprinting less as the tournament progressed). That’s must mean that players are good at recovering and have good fitness. The matches that seemed the most affected by fatigue were the second match and the last match. As discussed above, that is expected because the second match was played the day after the first match. All of the other matches had at least one day’s rest in between. The last match presumably had lower values due to accumulated fatigue. In conclusion, fatigue definitely has an effect on player’s physical performance, but this needs to be analyzed further, and it varies on a case-by-case basis. High performing athletes seem to be very good at recovering quickly and minimizing the effects of fatigue.
library(dplyr)# Assuming your dataframe is called df and the columns are named 'Player.Name' and 'Maximum.Velocity'max_speed_by_player <- master_df %>%group_by(Player.Name) %>%summarise(Max_Speed =max(Maximum.Velocity, na.rm =TRUE)) %>%ungroup()# Assuming 'Identifier' in name_mapping matches 'Player.Name' in max_speed_by_playermax_speed_with_names <- max_speed_by_player %>%left_join(name_mapping, by =c("Player.Name"="Identifier")) %>%select(Original_Name, Max_Speed)# View the resultsmax_speed_with_names
# A tibble: 35 × 2
Original_Name Max_Speed
<chr> <dbl>
1 Gabriel Arnold 17.9
2 Gustavo Gonzalez 18.7
3 Ryker Joutz 19.2
4 Allan Legaspi 20.6
5 Marcus Lightbourne 19.2
6 Julian Placias 20.7
7 Paulo Rudisill 19.5
8 Nico Schelotto 18.6
9 Harbor Miller 19.4
10 Nathan Nava 17.7
# ℹ 25 more rows
library(readxl)# Replace 'path_to_your_excel_file.xlsx' with the actual path to your Excel fileplayer_data <-read_excel("/Users/sylwialipior/Downloads/pm566-01-lab/Academy Data.xlsx")# Create a new 'Original_Name' column by concatenating 'First_Name' and 'Last_Name'player_data <- player_data %>%mutate(Original_Name =paste(First_Name, Last_Name))# Join the player data with max speed data# Make sure the key column names match those in your actual dataframescombined_data <- player_data %>%left_join(max_speed_with_names, by ='Original_Name')library(writexl)# Replace 'path_to_new_excel_file.xlsx' with the desired path for the new Excel filewrite_xlsx(combined_data, '/Users/sylwialipior/Downloads/pm566-01-lab/Academy Data_updated.xlsx')# Continuing from the previous combined_data dataframecombined_data <- combined_data %>%mutate(Height_m = Height /100, # Convert height from centimeters to metersWeight_kg = Weight *0.453592, # Convert weight from pounds to kilogramsBMI = Weight_kg / (Height_m^2) # Calculate BMI )# View the first few rows to confirm the BMI column has been addedhead(combined_data)
# A tibble: 6 × 22
Date Last_Name First_Name Birth_Year Birth_Month Gender Team
<dttm> <chr> <chr> <dbl> <chr> <chr> <chr>
1 2022-08-01 00:00:00 Dalgado Riley 2006 August M U17
2 2022-08-01 00:00:00 Garcia Emiliano 2006 January M U17
3 2022-08-01 00:00:00 Legaspi Allan 2006 January M U17
4 2022-08-01 00:00:00 Placias Julian 2006 April M U17
5 2022-08-01 00:00:00 Dunbar Adam 2007 February M U17
6 2022-08-01 00:00:00 Lightbourne Marcus 2006 July M U17
# ℹ 15 more variables: Height <dbl>, Weight <dbl>, Body_Fat <lgl>, Yo_Yo <dbl>,
# ten_meter <dbl>, `20m` <dbl>, thirty_meter <dbl>, five_ten_five_L <dbl>,
# five_ten_five_R <dbl>, CMJ <chr>, Original_Name <chr>, Max_Speed <dbl>,
# Height_m <dbl>, Weight_kg <dbl>, BMI <dbl>
library(dplyr)library(lubridate)
Attaching package: 'lubridate'
The following objects are masked from 'package:base':
date, intersect, setdiff, union
combined_data <- combined_data %>%mutate(# Convert 'Birth_Month' to a numeric valueBirth_Month_Num =match(Birth_Month, month.name), # Assumes 'Birth_Month' is a three-letter abbreviationBirthdate =make_date(Birth_Year, Birth_Month_Num, 1),# Make sure 'Date' is a Date object; use the appropriate lubridate function if the format is differentTest_Date =as.Date(Date), # Replace with mdy(Date), dmy(Date), etc., as appropriateAge_at_Testing =as.numeric(difftime(Test_Date, Birthdate, units ="weeks")) /52.25 )# View the first few rows to confirm the Age_at_Testing column has been addedhead(combined_data)
# A tibble: 6 × 26
Date Last_Name First_Name Birth_Year Birth_Month Gender Team
<dttm> <chr> <chr> <dbl> <chr> <chr> <chr>
1 2022-08-01 00:00:00 Dalgado Riley 2006 August M U17
2 2022-08-01 00:00:00 Garcia Emiliano 2006 January M U17
3 2022-08-01 00:00:00 Legaspi Allan 2006 January M U17
4 2022-08-01 00:00:00 Placias Julian 2006 April M U17
5 2022-08-01 00:00:00 Dunbar Adam 2007 February M U17
6 2022-08-01 00:00:00 Lightbourne Marcus 2006 July M U17
# ℹ 19 more variables: Height <dbl>, Weight <dbl>, Body_Fat <lgl>, Yo_Yo <dbl>,
# ten_meter <dbl>, `20m` <dbl>, thirty_meter <dbl>, five_ten_five_L <dbl>,
# five_ten_five_R <dbl>, CMJ <chr>, Original_Name <chr>, Max_Speed <dbl>,
# Height_m <dbl>, Weight_kg <dbl>, BMI <dbl>, Birth_Month_Num <int>,
# Birthdate <date>, Test_Date <date>, Age_at_Testing <dbl>
Linear Regression
# Let's assume your combined data with max speed, BMI, and age is named 'combined_data'model <-lm(Max_Speed ~ Age_at_Testing + Height_m + Weight_kg + thirty_meter + ten_meter, data = combined_data)summary(model)
Call:
lm(formula = Max_Speed ~ Age_at_Testing + Height_m + Weight_kg +
thirty_meter + ten_meter, data = combined_data)
Residuals:
Min 1Q Median 3Q Max
-0.9410 -0.4834 -0.1384 0.4963 1.0577
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 64.08743 11.33309 5.655 1.88e-05 ***
Age_at_Testing -0.34688 0.32666 -1.062 0.302
Height_m -1.65461 3.85506 -0.429 0.673
Weight_kg -0.02074 0.03461 -0.599 0.556
thirty_meter -8.17774 1.35102 -6.053 8.03e-06 ***
ten_meter -0.16515 0.89072 -0.185 0.855
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.6928 on 19 degrees of freedom
(140 observations deleted due to missingness)
Multiple R-squared: 0.7461, Adjusted R-squared: 0.6793
F-statistic: 11.17 on 5 and 19 DF, p-value: 3.963e-05
library(caret)
Loading required package: lattice
library(dplyr)# Filter out rows with NA in the thirty_meter columncombined_data_filtered <- combined_data %>%filter(!is.na(thirty_meter) &!is.na(Max_Speed) &!is.na(ten_meter))# Set seed for reproducibilityset.seed(123)# Split data into training (75%) and test sets (25%)trainIndex <-createDataPartition(combined_data_filtered$Max_Speed, p =0.8, list =FALSE, times =1)trainData <- combined_data_filtered[trainIndex, ]testData <- combined_data_filtered[-trainIndex, ]# Fit the linear model on the training setmodel <-lm(Max_Speed ~ Age_at_Testing + Height_m + Weight_kg + thirty_meter + ten_meter, data = trainData)# Summarize the modelsummary(model)
Call:
lm(formula = Max_Speed ~ Age_at_Testing + Height_m + Weight_kg +
thirty_meter + ten_meter, data = trainData)
Residuals:
Min 1Q Median 3Q Max
-1.0378 -0.3590 -0.1233 0.5099 1.0563
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 68.699448 14.266118 4.816 0.000227 ***
Age_at_Testing -0.412015 0.465008 -0.886 0.389581
Height_m -3.946296 4.228569 -0.933 0.365473
Weight_kg -0.006261 0.038711 -0.162 0.873675
thirty_meter -8.205729 1.545447 -5.310 8.74e-05 ***
ten_meter -0.397217 0.944680 -0.420 0.680098
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.7118 on 15 degrees of freedom
Multiple R-squared: 0.7681, Adjusted R-squared: 0.6907
F-statistic: 9.934 on 5 and 15 DF, p-value: 0.00024
# Predict on the test setpredictions <-predict(model, newdata = testData)# Calculate performance metrics, such as Mean Squared Error (MSE)mse <-mean((testData$Max_Speed - predictions)^2)print(paste("Mean Squared Error: ", mse))
[1] "Mean Squared Error: 0.446101767664014"
# Optionally, compare actual max speeds vs predicted max speedscomparison <-data.frame(Actual = testData$Max_Speed, Predicted = predictions)print(comparison)
The following object is masked from 'package:dplyr':
recode
# Fit the linear model without the 'ten_meter' variablemodel_without_ten <-lm(Max_Speed ~ Age_at_Testing + Height_m + Weight_kg + thirty_meter, data = trainData)# Calculate VIF for the modelvif_results <-vif(model_without_ten)print(vif_results)
